# SummQA at MEDIQA-Chat 2023:In-Context Learning with GPT-4 for Medical Summarization

**Authors:** Yash Mathur, Sanketh Rangreji, Raghav Kapoor, Medha Palavalli, Amanda Bertsch, Matthew R. Gormley  
**arXiv:** [2306.17384](https://arxiv.org/abs/2306.17384)  
**Date:** 2023-06-30  
**Summary page:** https://www.summarizepaper.com/en/paper/2306.17384v1/

## Key points

- Unstructured nature of medical conversations
- Use of medical terminology in gold summaries
- Identifying key information across multiple symptom sets
- Proposed system for Dialogue2Note Medical Summarization tasks:
- Two-stage process for section-wise summarization (Task A)
- Selecting semantically similar dialogues
- Using top-k similar dialogues as in-context examples for GPT-4
- Similar solution with k=1 for full-note summarization (Task B)
- 3rd place in Task A (2nd among all teams)
- 4th place in Task B Division Wise Summarization (2nd among all teams)
- 15th place in Task A Section Header Classification (9th among all teams)
- Overall, achieved 8th place in Task B
- Effectiveness of few-shot prompting for the task
- GPT-4 summaries are more abstractive and shorter
- Code made publicly available for further research and development
- Innovative approach using GPT-4 for both section-wise and full-note summarization tasks
- Contributions to the field and insights into strengths and weaknesses of prompting-based approaches.

## Summary

In their paper titled "SummQA at MEDIQA-Chat 2023: In-Context Learning with GPT-4 for Medical Summarization," authors Yash Mathur, Sanketh Rangreji, Raghav Kapoor, Medha Palavalli, Amanda Bertsch, and Matthew R. Gormley address the challenges of medical dialogue summarization. They highlight the unstructured nature of medical conversations, the use of medical terminology in gold summaries, and the need to identify key information across multiple symptom sets. To tackle these challenges, the authors propose a novel system for the Dialogue2Note Medical Summarization tasks in the MEDIQA 2023 Shared Task. For section-wise summarization (Task A), they employ a two-stage process that involves selecting semantically similar dialogues and using the top-k similar dialogues as in-context examples for GPT-4. For full-note summarization (Task B), they adopt a similar solution with k=1. The authors achieved impressive results in the shared task, securing 3rd place in Task A (2nd among all teams), 4th place in Task B Division Wise Summarization (2nd among all teams), 15th place in Task A Section Header Classification (9th among all teams), and 8th place overall in Task B. Their success highlights the effectiveness of few-shot prompting for this task. However, they also acknowledge some weaknesses associated with prompting-based approaches. In their evaluation, the authors compare GPT-4 performance with several finetuned baselines and observe that GPT-4 summaries are more abstractive and shorter. They make their code publicly available to facilitate further research and development in this area. Overall, this paper provides valuable insights into medical dialogue summarization and presents an innovative approach using GPT-4 for both section-wise and full-note summarization tasks. The authors' achievements and findings contribute to the advancement of this field, while also shedding light on the strengths and weaknesses of prompting based approaches.